Does Intel/AMD check their cpu's

[Marvzl1357]

Hullo,

Doing my maths homework, and something came up about cpu's, exact wording is "In the mass production of computer processors it is found that 5% are defective. Processors are selected at random and put into packs of 15." then find probability of yadayada.

Anyway, this sparked a thought, 5%? is that all? I thought it might be more. This is a maths textbook though, they're probably just fudging these numbers anyway.

Then the second part, the "selecting at random" part, I doubt Intel or AMD would simply grab any random ones off the construction line, surely they test them?

Anyway, I'm tired, bored of maths, and my mind is anywhere but maths at the moment.

Anyone actually know what the manufacturing yield is? And do they check every cpu that they produce or not?

they should have made the question 5% of 4770K's can be overclocked to 4.8 at a voltage of less than 1.3v, find the probability that, out of 15, at least three processors can achieve this.

I joke, I'm bored

[Enderman]

How old is your math textbook? The failure rates are much lower than that nowadays...

The manufacturers check each CPU using imaging software to make sure that the silicon chip is undamaged.

Then they take a certain amount out of every batch for testing, not every single one. That would be way too much work to do, since they produce millions of processors per year.

[Geekazoid]

Hullo,

Doing my maths homework, and something came up about cpu's, exact wording is "In the mass production of computer processors it is found that 5% are defective. Processors are selected at random and put into packs of 15." then find probability of yadayada.

Anyway, this sparked a thought, 5%? is that all? I thought it might be more. This is a maths textbook though, they're probably just fudging these numbers anyway.

Then the second part, the "selecting at random" part, I doubt Intel or AMD would simply grab any random ones off the construction line, surely they test them?

Anyway, I'm tired, bored of maths, and my mind is anywhere but maths at the moment.

Anyone actually know what the manufacturing yield is? And do they check every cpu that they produce or not?

That's ridiculous!

[Izaya Orihara]

It's more like 1 these days. 

[Sauron]

Your book is surely inventing these numbers, I don't think they would bother doing actual research on the subject when the only reason it talks about cpus is to give a bit of context to problems which would otherwise be always equal.

 

As for testing, cpus are not manifactured by intel and amd themselves but by global foundries, who do their own quality controls. They can't possible stress test every single cpu, they check if it fits the requirements on paper and proceed to ship them. Once in a while they pick one at random from the production line and actuall test it. Depending on how many defective ones they find they decide if they need to revise the production process or not. And by the way i5's are actually failed i7's which would be unstable if hyperthreading was turned on, just like fx 6xxx cpus are 8xxx cpus with one piledriver block that doesn't work properly and gets fused off.

[LukaP]

ill explain on an example. 4770k is a non defective CPU, as it has everything enabled, that that die has built in (looking at the consumer side, i know they purpously disable ECC and some VT things from xeons)

 

then you go to a 4770 and 4771 that have a locked multi, since they probably cant go any higher anyway. that is the first defect

 

then 4670k has HT disabled, since atleast in one core, it is defective

 

4670 is to 4670k what a 4770 is to a 4770k

 

then you have lower clocked i5s, these are usually bad silicon, that cant clock that high.

 

 

then you get to the defects that make a chip into an i3. (either 1/2 cores dead, or 1 core dead, one cant be HTd)

these have lower clocked variants too

 

then even more defects means not even the 2 cores that are ok can hyperthread, this means you get a pentium from the die.

 

and if enough cache or 3 cores are dead, you get a celeron. 

 

and if all 4 cores are dead, (which im expecting less than 5% of the dies are) then you have a full defect

[ZeroBreaker]

Actually, there was a bbc documentary from the 1970's that went over the applications of microprocessors in the future that was reshown on iplayer recently. Before silicon took off, it was lucky if there was a 50% failure rate - one speck of dust in the factory could lead to the whole factory being rendered useless, no exaggeration. It's remarkable really just how complicated this process is.

Back then, they put a dot of die or some such material to check if it worked (somehow).

Heres the link for anyone in the UK :

http://www.bbc.co.uk/iplayer/episode/p01z4rrj/horizon-19771978-now-the-chips-are-down

Youtube link at lower quality but global:

https://www.youtube.com/watch?v=ll_-_ngu4Gg

It's a really interesting documentary.

As for modern times, considering one bent chip could kill a cpu, i would suggest they do check every chip - otherwise returns would be extremely frequent, and despite my many cpu purchases for personal use and for other people, i have never had a single bad cpu.

[Createinator]

-snip-

Thank you, that was very informative!

[LukaP]

Thank you, that was very informative!

any time. and the same applies to intels 2011 lineup and their Xeons. they basically (as far as the info ive had confurms) have 3 different dies. a mainstream 4 haswell cores, an enthusiast 6 Ivy cores (soon to be 8 haswell cores) and an enterprise 15 haswell cores and then they just scale it down on different CPUs according to how much of the die actually works

[Marvzl1357]

ill explain on an example. 4770k is a non defective CPU, as it has everything enabled, that that die has built in (looking at the consumer side, i know they purpously disable ECC and some VT things from xeons)

 

then you go to a 4770 and 4771 that have a locked multi, since they probably cant go any higher anyway. that is the first defect

 

then 4670k has HT disabled, since atleast in one core, it is defective

 

4670 is to 4670k what a 4770 is to a 4770k

 

then you have lower clocked i5s, these are usually bad silicon, that cant clock that high.

 

 

then you get to the defects that make a chip into an i3. (either 1/2 cores dead, or 1 core dead, one cant be HTd)

these have lower clocked variants too

 

then even more defects means not even the 2 cores that are ok can hyperthread, this means you get a pentium from the die.

 

and if enough cache or 3 cores are dead, you get a celeron. 

 

and if all 4 cores are dead, (which im expecting less than 5% of the dies are) then you have a full defect

Nice

 

How old is your math textbook? The failure rates are much lower than that nowadays...

The manufacturers check each CPU using imaging software to make sure that the silicon chip is undamaged.

Then they take a certain amount out of every batch for testing, not every single one. That would be way too much work to do, since they produce millions of processors per year.

From oxford printing press 2012.

I sorta knew that they did a random sample, but just double checking

 

Your book is surely inventing these numbers, I don't think they would bother doing actual research on the subject when the only reason it talks about cpus is to give a bit of context to problems which would otherwise be always equal.

 

As for testing, cpus are not manifactured by intel and amd themselves but by global foundries, who do their own quality controls. They can't possible stress test every single cpu, they check if it fits the requirements on paper and proceed to ship them. Once in a while they pick one at random from the production line and actuall test it. Depending on how many defective ones they find they decide if they need to revise the production process or not. And by the way i5's are actually failed i7's which would be unstable if hyperthreading was turned on, just like fx 6xxx cpus are 8xxx cpus with one piledriver block that doesn't work properly and gets fused off.

Yeah, it definitely is making up these numbers, I was bored, and when you find something that interests you when you're bored, you fly with it.

I wasn't exactly sure who made the actual chips. So Intel designs them, then sends the "blueprints" to global foundries who then manufactures them to those specs?

And the textbook didn't define what a failure is.

 

Geez, I'm over thinking this...

[LukaP]

Nice

 

From oxford printing press 2012.

I sorta knew that they did a random sample, but just double checking

 

Yeah, it definitely is making up these numbers, I was bored, and when you find something that interests you when you're bored, you fly with it.

I wasn't exactly sure who made the actual chips. So Intel designs them, then sends the "blueprints" to global foundries who then manufactures them to those specs?

And the textbook didn't define what a failure is.

 

Geez, I'm over thinking this...

intel has its own fabs they design and print themselves. thats why they are at 14nm while AMD is struggling to get 22FF goign

[Sauron]

Yeah, it definitely is making up these numbers, I was bored, and when you find something that interests you when you're bored, you fly with it.

I wasn't exactly sure who made the actual chips. So Intel designs them, then sends the "blueprints" to global foundries who then manufactures them to those specs?

Sorry actually global foundries doesn't manifacture intel products I got that wrong. It's nvidia and amd that do that

[Marvzl1357]

Sorry actually global foundries doesn't manifacture intel products I got that wrong. It's nvidia and amd that do that

oh, ok. Both amd and nvidia use global foundries?

 

intel has its own fabs they design and print themselves. thats why they are at 14nm while AMD is struggling to get 22FF goign

So, Intel and AMD have their own fabs? they make their own cpu's?

[LukaP]

oh, ok. Both amd and nvidia use global foundries?

 

So, Intel and AMD have their own fabs? they make their own cpu's?

nvidia uses TSMC. amd uses TSMC for GPUs and GF for CPUs. Intel has its own fabs for everything